ERROR:__main__:Error processing batch: (sqlalchemy.dialects.postgresql.asyncpg.Error) <class 'asyncpg.exceptions.StringDataLengthMismatchError'>: bit string l
ength 536 does not match type bit(64)                                                                                                                         
[SQL: UPDATE messages SET message_hash=$1, updated_at=now() WHERE messages.id = $2::BIGINT]                                                 
[parameters: [(b"B'0000100101000100000101001100001100011011010111000111101111111110'", 23189), (b"B'1000100001000100000101001100011100010011001110001101101111
011110'", 724950), (b"B'1011011100000100010001111110101111001011111011010110101111011010'", 103289), (b"B'1001101100010100010001100100001101011011001111011100
101111111010'", 90)]]                                                                                                                                         
(Background on this error at: https://sqlalche.me/e/20/dbapi)                                                                                                 
ERROR:__main__:Fatal error: (sqlalchemy.dialects.postgresql.asyncpg.Error) <class 'asyncpg.exceptions.StringDataLengthMismatchError'>: bit string length 536 d
oes not match type bit(64)

sqlalchemy.exc.DBAPIError: (sqlalchemy.dialects.postgresql.asyncpg.Error) <class 'asyncpg.exceptions.StringDataLengthMismatchError'>: bit string length 528 does not match type bit(64)
[SQL: UPDATE messages SET message_hash=$1, updated_at=now() WHERE messages.id = $2::BIGINT]
[parameters: [(b"'0000100101000100000101001100001100011011010111000111101111111110'", 23189), (b"'1000100001000100000101001100011100010011001110001101101111011110'", 724950), (b"'1011011100000100010001111110101111001011111011010110101111011010'", 103289), (b"'1001101100010100010001100100001101011011001111011100101111111010'", 90)]]
(Background on this error at: https://sqlalche.me/e/20/dbapi)

from pgvector.sqlalchemy import BIT
from pgvector.utils import Bit

class Messages(Base):
    __tablename__ = "messages"
    id: Mapped[int] = mapped_column(BigInteger, primary_key=True)

    # other columns...

    message_hash: Mapped[np.ndarray | None] = mapped_column(BIT(64), nullable=True)

assert embeddings_hash.ndim == 1, f"{embeddings_hash.shape=}"
assert embeddings_hash.shape[0] == len(batch), f"{embeddings_hash.shape=} {len(batch)=}"
assert embeddings_hash.dtype == np.uint64, f"{embeddings_hash.dtype=}"

embeddings_hash = embeddings_hash.view(np.uint8)  # shape = (batch_size * 8,)
embeddings_hash = embeddings_hash.reshape(configuration["batch_size"], 8)
embeddings_hash = np.unpackbits(embeddings_hash, axis=1) 

updates = [
    {
        "b_id": message.id,
        "message_hash": f"B'{Bit(embedding_hash.reshape(-1)).to_text()}'".encode("ascii")
    }
    for message, embedding_hash in zip(batch, embeddings_hash)
]

await session.execute(
    Messages.__table__.update()
    .where(Messages.id == bindparam("b_id"))
    .values({"message_hash": bindparam("message_hash")})
    .execution_options(synchronize_session=False),
    updates
)